The individual tree segmentation is of great significance for forest resource surveys. The accuracy of segmentation is profoundly influenced by the choice of the forest single-wood segmentation algorithm and the parameters associated with different structural complexities. This research introduced drone orthophoto and laser radar data from Tianheng Island. Initially, 2D and 3D characteristics of typical forest vegetation were extracted. Subsequently, the random forest algorithm was applied to classify different tree species. With the classification of point cloud data, sampling plots with varying structural complexities were selected to conduct comparative analysis encompassing clustering algorithms, stacking algorithms, and the watershed algorithm, in order to enhance segmentation accuracy. The findings reveal that: (1) the random forest algorithm, combined with 2D and 3D features of a single wood, effectively classifies the mixed forest trees, achieving an impressive overall accuracy of 94.51% and a Kappa coefficient of 0.9038. (2) The clustering algorithm shows the highest segmentation accuracy for forest areas with simple structures (F=96.41), while depending upon the selection of segmentation parameters. In the case of complex single wood clusters, the watershed algorithm displays the least fluctuation (ΔF=14.56), indicating its superior stability. (3) Pre-classification of tree species in mixed forests effectively improves the single wood segmentation environment. Compared to direct single wood segmentation, the clustering, stacking, and watershed algorithms yield increased segmentation accuracy to varying degrees (ΔF₁=10.06, ΔF₂=9.51, ΔF₃=12.6).